Diffuse large B-cell-lymphoma (DLBCL) is a heterogeneous disease in its sites of origin, genetic alterations, and clinical behavior. Gene expression profiling (GEP) has led to the identification of two molecular subtypes, GCB-like and ABC-like DLBCL, that follow different molecular circuits and hence, likely represent different diseases. Next to the ABC-like GEP, the rearrangement and/or expression of MYC and TP53 mutations in tumors characterize patient subsets with inferior prognosis. However, the clinical impact of cell of origin (COO) subtyping and the identification of prognostic biomarkers differ between studies. In many clinical trials, patients identified "at risk" experience cure and long-time survival. Important factors modulating the impact of tumor cell-specific factors may be conferred by the host microenvironment, and stromal signatures have been shown to be correlated to survival in DLBCL. The robust analysis of stromal signatures is hampered by the lack of assays applicable to routinely used formalin-fixed paraffin-embedded (FFPE) materials. We have constructed a molecular signature applicable to FFPE, interrogating the quantitative and qualitative composition of the microenvironment in DLBCL. The signature was trained using an algorithm that extracts prognostic information out of the ratios of pairs of genes. The genes that drive prognosis are expressed in T-cells and macrophages and have function in the communication between both cell types. The model was validated using the NanoString assay in a cohort of 466 DLBCL patients enrolled in 7 prospective clinical trials (MInT, MegaCHOEP phase III and observation, RICOVER-60, RICOVER-noRTh, DENSER, SMARTER) of the German High grade non-Hodgkin's lymphoma study group (DSHNHL). Grouping of the patients into quartiles according to the expression of the continuous stromal signature score (ranging from -1.880to 4.441) resulted in three quartiles (Q1-3) with comparable clinical behavior (stromal signature low). Patients from quartile 4 (Q4), however, characterized by high expression of the signature (stromal signature high), showed a clearly inferior outcome in EFS, PFS and OS (Figure 1 a-c). This result could be independently reproduced in the seven clinical trials named above, thus clearly depicting the robustness of the signature. Multivariate analysis revealed that high expression of the stromal signature is a prognostic risk factor independent of the IPI factors in EFS (HR 1.7, 95 % CI 1.2-2.4, p-value =0.001), PFS (HR 1.8, 95 % CI 1.2-2.5, p-value =0.001) and OS (HR1.8, 95 % CI 1.3-2.7, p-value =0.001). Combining stromal signature count with IPI-score led to the identification of a high-risk cohort (Fig. 1 d-f). Of importance, additional multivariate analyses adjusted for the IPI factors performed within selected trials showed that the stromal signature provides prognostic information independent of the COO status, MYC and dual MYC/BCL2 rearrangements, TP53 mutations and the MYC/BCL2 double expresser status.

In summary, our data from prospectively randomized trials of the DSHNHL underline the importance of the microenvironment in the prognostic stratification of DLBCL patients and suggest that the composition and quality of the tumor stroma is an independent risk factor in DLBCL. Analysis of stromal features, therefore, may provide a rationale for targeted treatment approaches, e.g. with immunomodulatory substances (IMIDs), in patients at risk.


Klapper:Regeneron: Honoraria, Research Funding; HTG Molecular Diagnostics, Inc.: Research Funding; F.Hoffman-La Roche: Honoraria, Research Funding; Amgen: Honoraria, Research Funding; Takeda: Honoraria, Research Funding. Richter:HTG Molecular Diagnostics, Inc.: Research Funding. Poeschel:Roche: Other: Travel grants; Amgen: Other: Travel grants. Held:BMS: Consultancy, Other: Travel grants, Research Funding; Spectrum: Research Funding; Roche: Consultancy, Other: Travel grants, Research Funding; Amgen: Research Funding; MSD: Consultancy.

Author notes


Asterisk with author names denotes non-ASH members.